A perceptual system for vision-based evolutionary robotics

摘要

This paper discusses the perceptual system of a mobile robot from the viewpoint of ecological psychology. The control of the mobile robot is based on the concept of perceiving-acting cycle. We discuss the perceptual system for extracting perceptual information that will be used and required for next action. Based on action outputs, the perceptual module is selected and accommodated in order to persist a specific perception. Furthermore, we apply a steady-state genetic algorithm (SSGA) for extracting perceptual information from an image. The average and divergence used for mutation in the SSGA for detecting a pattern are updated according to the accuracy of detection. The current searching result of SSGA is reflected to the next search of SSGA. In this way, the searching process of the perceptual system for the mobile robot is cyclic according to the spatio-temporal context of the environment. We apply the proposed method for the control of a mobile robot used under an office automation floor. Experimental results show the effectiveness of the proposed method.